METHODOLOGY · AUTONOMOUS RISK INTELLIGENCE

How Fahali sees
what markets hide.

Eighteen detection layers. Seven-engine consensus. Sub-300ms from tick to decision. This is the architecture behind the most accessible institutional-grade risk surveillance platform ever built.

18
Detection layers
9,142
Instruments monitored
5
Asset classes
<300ms
Tick to decision
72h
Avg. forecast horizon

The detection pipeline

Every tick of every instrument passes through Fahali's detection pipeline in under 300 milliseconds. The pipeline has five stages, each adding a layer of confidence before an alert reaches your phone.

+12ms
Ingest
Raw market data from exchange feeds
+38ms
Detect
18 detection layers fire
+18ms
Confirm
7-engine consensus validation
+8ms
Narrate
Plain English reasoning trace
+2ms
Deliver
Push to your dashboard or API

The 18 detection layers

Each layer applies a distinct statistical or machine-learning method to detect a specific class of market anomaly. No single layer decides alone — anomalies must survive the consensus layer.

01

Volatility

IV/RV divergence, term structure backwardation, volatility regime flips (2.6x+ expansion detected as regime change).

02

Momentum

Multi-timeframe momentum divergence, ROC-based shift detection across 5 time horizons ranging from 15 minutes to 4 hours.

03

Dark Pool

Off-exchange block trade detection, stealth accumulation, exchange dark pool volume anomalies against rolling median baselines.

04

Liquidity

Order book thinning detection, liquidity vacuum identification (volume dropping below 5% of median), liquidity wall mapping at key price levels.

05

Order Imbalance

Bid/ask pressure divergence, aggressive vs passive order flow ratio, cumulative delta analysis on per-tick basis.

06

Regime Detection

Hidden Markov Model (HMM) classifying 4 market regimes — Bullish, Bearish, Neutral, High Volatility — with confidence scores and diag covariance.

07

Divergence

Intermarket correlation break detection, cross-asset coupling strength monitoring, rolling window correlation with Z-score thresholding.

08

Magnet

Options gamma level detection, concentration mapping at strike prices with high open interest, max pain level identification.

09

Confluence

Multi-indicator confluence detection — when 3+ independent layers fire simultaneously, confidence is automatically escalated.

10

Volume Spike

Statistical volume anomaly detection using rolling Z-score with 4.0+ standard deviation threshold, regime-aware baselines.

11

Reversal

Wyckoff pattern recognition, climax volume identification, exhaustion flagging with capped confidence for directional reversal signals.

12

Stealth

Institutional accumulation pattern detection — price-neutral volume accumulation, iceberg order inference from L2 data.

13

Correlation

Cross-asset correlation structure break detection, tail-dependence matrix computation, contagion path simulation through the correlation graph.

14

Flow

Dark pool + derivatives + on-chain flow triangulation, net institutional flow estimation, OTC settlement inference.

15

Squeeze

Leverage cascade modeling, margin call waterfall simulation, liquidation cascade path estimation across correlated assets.

16

Funding

Funding rate divergence detection, perpetual futures basis monitoring, cross-exchange funding arbitrage flagging.

17

Skew

Options skew monitoring, risk reversal detection, tail risk pricing analysis across major derivatives venues.

18

Exhaustion

Volume-weighted delta divergence, high-conviction exhaustion pattern detection with directionless flag for pure volatility expansion signals.

Capital flow reconstruction

Fahali reconstructs institutional capital movement from three independent data sources, triangulated to estimate net flow direction and magnitude in near-real-time. This is the flow surface — the second of three institutional-grade surfaces.

Layer 1: Dark Pool

Off-exchange block trades and hidden liquidity that never reaches public order books. Fahali monitors L2 data across major ATS/ dark pool venues, inferring block trades from tape-read anomalies and odd-lot patterns.

Layer 2: Derivatives Flow

CME futures, Deribit options, and perpetual swap markets reveal directional positioning that spot markets lag. Fahali computes rolling net delta, open interest divergence, and basis signals across venues.

Layer 3: On-Chain Settlement

Exchange wallet outflows, smart contract interactions, and OTC settlement patterns are extracted from blockchain data and correlated with L2 and derivatives signals to confirm flow direction.

Net 24h flow estimate: +$1.65B institutional across tracked venues (representative). Dark pool +$847M. Derivatives +$612M. On-chain +$284M. Spot/public retail −$94M. Updated every 60 seconds.

Contagion mapping

Fahali models cross-asset correlation as a dynamic graph — when stress flares in one node, the platform simulates which holdings absorb the shock, in what order, and at what timing. This is the third institutional-grade surface.

The contagion model uses a tail-dependence matrix computed over rolling windows with regime-aware coupling strength. Unlike simple correlation matrices, tail-dependence captures nonlinear co-movement during stress events — the relationships that only appear when markets are falling.

Key methodology

Accuracy and transparency

Fahali publishes its detection accuracy publicly. Every signal is tracked, every outcome is resolved against subsequent price action, and every miss is published. This is not optional — it is the foundation of institutional trust.

88%
Magnitude precision (blended)
80.5%
Crash detection precision
+35pp
Average base-rate lift
291M
Labeled detection corpus

See the full accuracy scorecard for per-engine breakdowns, base-rate comparisons, and methodology.

The number that matters is not the raw percentage — it is the base-rate lift. In a one-directional market (all assets falling), a naive "always bearish" strategy would score 100% on direction. Fahali's 88% magnitude precision represents a 33-45 percentage point improvement over the naive base rate, depending on the engine. The scorecard is designed to measure genuine skill, not vanity metrics.

Frequently asked questions

How does Fahali detect market anomalies?

Fahali runs 18 statistical detection layers across every tick of every instrument. Each layer applies a different method — volatility regime detection, HMM regime classification, dark pool volume analysis, order flow imbalance, leverage cascade modeling, and more. Anomalies are confirmed by a 7-engine consensus before any alert is generated.

How far in advance does Fahali predict market stress?

Signals are issued up to 72 hours in advance, depending on the engine. Every forecast includes a confidence interval and a named set of drivers so you can evaluate it against your own judgment.

What data sources does Fahali use?

Direct exchange feeds (CME, Cboe, Deribit, Coinbase Prime, Binance), Level 2 order books, on-chain settlement data from major blockchain networks, and derivatives flow data. Fahali does not use scraped data or third-party data aggregators.

Is Fahali accurate?

Magnitude detection precision averages 88% across all engines with a base-rate lift of 33-45 percentage points. Crash detection precision is 80.5% against a bearish base rate of 42.6%. Every miss is published in the public scorecard. See the full accuracy page for details.

See the architecture in action.

14-day trial · No card required · Full access to all 18 layers